基于GRNN的导航卫星钟差仿真方法研究  

Study on Simulating Method of Navigation Satellite Clock Error Using Generalized Regression Neural Network

作　　者：王威[1] 荣志飞[1] 范国清[1] 

机构地区：[1]国防科技大学机电工程与自动化学院,长沙410073

出　　处：《系统仿真学报》2011年第11期2339-2345,共7页Journal of System Simulation

基　　金：国家863计划(2007AA12Z307)

摘　　要：提出了一种基于广义回归神经网络(GRNN)的导航卫星钟差仿真的新方法,根据IGS提供的精密卫星钟差序列建立基于广义回归神经网络的钟差模型。结合仿真实例,详细讨论了训练样本的采样时间间隔、输入维数和模型平滑因子对网络模型性能的影响,并确定了较优的采样时间间隔、输入维数和平滑因子。通过与常用的二次多项式模型进行对比分析,证明了GRNN模型在24h的预报时间跨度内精度仍可达ns级,初步验证了将GRNN模型用于钟差仿真的可行性和较多项式模型更优的实用性。A new method for the simulating of Navigation Satellite clock error was proposed based on generalized regression neural network. Firstly, according to the precise satellite clock error series from IGS, the model was established. Then, it was discussed that how the sampling interval of training samples, the dimension of input vector and the smooth factor influence the performance of network model, and an optimal group of parameters were determined by simulation. Finally, the results from the comparison between the GRNN model and quadratic polynomial model reveal that the model accuracy of GRNN is higher than that of polynomial model, which is up to ns magnitude for 24-hour prediction span. It is preliminarily proved that it is feasible and more practical to apply GRNN on the simulating of clock error.

关 键 词：原子钟差 广义回归神经网络 二次多项式 模型精度 

分 类 号：TP391.9[自动化与计算机技术—计算机应用技术]